The performance of the 12-item General Health Questionnaire (GHQ) was tested against a modified version of a structured diagnostic interview for making DSM III-R diagnoses, the Composite International Diagnostic Interview (CIDI), in a population of 787 primary care patients, 214 of whom were interviewed. The Yoruba versions of both instruments were found to be feasible in this setting and the inter-rater reliability of the CIDI was good. The GHQ-12 showed a sensitivity of 68% and a specificity of 70% when only cases meeting the criteria for specific DSM III-R disorders were considered. The sensitivity was better for certain individual disorders and the overall performance was marginally improved when scoring was made according to the revised method proposed by Goodchild and Duncan-Jones.